Multi-trait genomic-enabled prediction enhances accuracy in multi-year wheat breeding trials

Area of Science:

• Agricultural Science
• Genetics
• Quantitative Genetics

Background:

• Genomic selection (GS) utilizes genomic information to predict breeding values.
• Accurate evaluation of prediction accuracy is essential for implementing genomic-based prediction models.
• Multi-trait data analysis offers potential for enhanced prediction accuracy in genomic selection.

Purpose of the Study:

• To compare prediction accuracy using different genomic prediction models with multi-trait data.
• To evaluate various measures for assessing prediction accuracy in genomic selection.
• To investigate the benefits of multi-trait versus single-trait models for predicting wheat quality and grain yield.

Main Methods:

• Utilized six large multi-trait wheat datasets (quality and grain yield).
• Employed four different prediction models to assess prediction accuracy.
• Compared conventional Pearson's correlation with a corrected Pearson's correlation derived from a bivariate model.

Main Results:

• Conventional Pearson's correlation underestimated true prediction accuracy.
• Corrected Pearson's correlation, derived from bivariate models, showed higher accuracy (2.53-11.46%) compared to other methods.
• Multi-trait models outperformed single-trait models for grain yield prediction (5.80-14.01% increase in accuracy).

Conclusions:

• Corrected Pearson's correlation provides a more reliable measure of prediction accuracy in genomic selection.
• Multi-trait genomic prediction models offer substantial benefits, particularly for traits with high genetic correlations.
• The study highlights the advantage of multi-trait approaches for improving genomic-enabled prediction accuracy in wheat breeding.